Browser QA, accessibility, and performance closeout

[iainherd]

Goal

Verify the site in real browser conditions before client review.

Source artefacts

• Lir_Website_PRD_v0.1-2.md §7–§8 and §11
• Lir_Website_Design_Language_Spec_v0.2.md §8–§9

Acceptance criteria

• Desktop and mobile screenshots reviewed.
• Console errors checked.
• Keyboard and reduced-motion behaviour checked.
• Lighthouse/performance budget checked or equivalent documented.
• Password gate is re-tested in unauthenticated and authenticated states.
• Findings are filed as follow-up issues or closed with evidence.

[iainherd]

Agent allocation

Primary profile: agent:browser-qa-reviewer
Required reviewer profiles: agent:design-director, agent:code-security-reviewer, agent:devops-deployer

Reason: Allocation reviewed after independent subagent analysis; profile set reflects primary ownership and required verification gates.

Orchestration rules:

• Morwenna/orchestrator retains final repo, PR, deployment, and closeout verification.
• Use fresh bounded subagents for implementation/review slices; do not ask child agents to ingest the full artefact set unless their issue explicitly requires it.
• Spec/brief/content/design compliance review happens before code-quality/security review.
• Subagent reports are evidence, not proof: parent must inspect diff/status and rerun relevant checks before closure.
• Spawn the named Hermes profile only for long-running independent work; otherwise use delegate_task with a self-contained, profile-specific prompt.

[iainherd]

Design critique gate

This issue must use the agency-grade critique workflow from #21 once landed:

• docs/design/critique-rubric.md

For major visual work, closeout requires:

• at least two screenshot critique passes;
• mobile and desktop evidence;
• explicit answer to “Does this look AI-generated?”;
• weakest element removed/refined before final approval;
• neutral placeholders only in pre-prod, no stock/AI stand-ins.